Alcohol Consumption Over Time and Risk of Death: A Systematic Review and Meta-analysis

Harindra Jayasekara; Dallas R. English; Robin Room; Robert J. MacInnis


Am J Epidemiol. 2014;179(9):1049-1059. 

In This Article


We summarized existing evidence from cohort studies for the association between alcohol consumption measured over time and the risk of death from all causes. A dose-response meta-analysis for men showed a nonlinear relationship with weak evidence for an inverse association at lower intake levels and an increased mortality risk for intakes over 40 g/day compared with abstention. An analysis of categorical data that was ancillary to the dose-response meta-analysis gave a 10% lower mortality risk for an intake of 1–29 g/day over time compared with abstention and an elevated mortality risk that was not statistically significant for an intake of 60 or more g/day over time for men. Heterogeneity between studies was present, as shown by the results for the I 2 statistic, but publication bias was not evident. Few studies reported relative risks for heavy drinking, and few reported results for women.

The complexity of the alcohol–heart disease relationship goes beyond current drinking and related possible biological mechanisms, and the usefulness of repeated measures of exposure has been highlighted.[35] Many of the included studies had a considerable time interval between measurements of exposure, thus allowing adequate time for consumption to vary over time in participants. In addition, because these were prospective cohort studies, recall bias was avoided.

The present meta-analysis also had several potential limitations. First, there was substantial heterogeneity between studies. Although this could be considered a limitation in terms of the meta-analysis, the examination of heterogeneity between studies was a key aim of the present paper. Alcohol is a complex product that is available as different beverage types (commonly beer, wine, and spirits), contains varying strengths of ethanol, and is offered in containers of different sizes, shapes, and names. Also, participants were usually asked to report their consumption using "standard drinks," the size of which varies from country to country.[12] The present systematic review also highlights the differences between studies in the methods used to classify consumption. Whereas Lazarus et al.,[23] Fillmore et al.,[25] Gronbaek et al.,[27] Emberson et al.,[28] and Friesema et al.[32] used intake categories based on drinks per unit time, Wellmann et al.,[26] Beulens et al.,[29] and Britton et al.[33] used grams per day, and Goldberg et al.[24] used milliliters per day. The cutoffs used to define light, moderate, and heavy drinking also varied among studies. In addition, the periods of follow-up also varied (range, 5–20 years). Considering the above, heterogeneity among studies seems inevitable. Even though Egger's test did not indicate evidence of publication bias in the analyses for men, the presence of heterogeneity makes cautious interpretation of results necessary.

Second, a meta-analysis cannot address problems with confounding that could be inherent in the original studies despite most authors having adjusted for key potential confounders. The lower mortality risk for consumption of less than 30 g/day compared with nondrinking could potentially be caused by residual confounding from hitherto undetected healthy behaviors related to low levels of drinking. Third, misclassification of alcohol consumption is common in self-reported observational studies of alcohol consumption. Bias usually arises from underreporting of alcohol consumption,[36,37] and this may have diminished the true inverse association between mortality risk and light drinking in the present analysis. Alcohol consumption could also have varied during follow-up, leading to misclassification. In the present study, reporting of methods and results was inconsistent across individual studies and, in some cases, made it difficult to interpret and recalculate reported relative risks. Another problem related to misclassification stems from the use of abstainers as the reference group without separating former drinkers, which did not allow us to completely eliminate the effect of those who have quit drinking because of illness.[7,38] Although at the level of a meta-analysis there is nothing that can be done to rectify this, most included studies had identified former drinkers. Fourth, the present study does not address the role of age in the association between alcohol mortality risk. Finally, the measure of exposure in the present dose-response and pooled meta-analysis does not reflect the pattern of drinking over time, which can be an important determinant of mortality risk.[39–41]

Meta-analyses of cohort studies of alcohol consumption measured at 1 time (baseline) only and mortality risk have previously reported a J-shaped curve.[3,4] English et al.,[4] in their meta-analysis based on 16 cohort studies, observed a 16% reduction in mortality risk at 1–2 drinks per day for men (RR = 0.84, 95% CI: 0.82, 0.86) and a 12% reduction in risk for up to 1 drink per day for women (RR = 0.88, 95% CI: 0.86, 0.90). English et al.[4] also observed that men consuming 4 or more drinks per day and women consuming 2 or more drinks per day had higher mortality risks than nondrinkers. Di Castelnuovo et al.[3] pooled 34 studies with a weighted regression analysis of fractional polynomials and found a significant reduction in mortality risk for up to 3 drinks per day for men and up to 2 drinks per day for women, whereas higher levels of alcohol intake were associated with higher mortality risk. The maximum reductions in risk (17% and 18% for men and women, respectively) were observed at alcohol intake levels below 10 g per day. The overall shape of the dose-response curve (P for nonlinearity = 0.02) and a higher mortality risk for heavier drinking for men in our analysis align with these findings in general, whereas the strength of the inverse association at lower levels of intake is weaker. Whether this is due to the currently available number of studies and their heterogeneity, or whether it is due to a true difference in the association between mortality risk and low alcohol intake when consumption is measured over time rather than at 1 time only (baseline in a cohort study) is not yet clear.

Studies encompassing measures of drinking over time should yield a closer approximation of theorized lifetime cumulative exposure to alcohol than studies using only baseline consumption.[6] Any net survival gain from low-dose exposure to alcohol is thought to act primarily through its "cardioprotective" effect, and the dose-response curve for death from all causes is determined by the relative frequency of ischemic heart disease in a population.[42] Biochemical evidence supports the existence of a causal relationship between regular low-dose drinking and protection from ischemic heart disease, potentially mediated via an increase in high-density lipoproteins, inhibition of platelet activation, reduction in fibrinogen levels, and production of antiinflammatory effects.[40,43,44] Moderate alcohol consumption has also been linked with a reduced risk of type 2 diabetes mediated via enhanced insulin sensitivity,[45] the production of metabolites such as acetaldehyde and acetate by ethanol oxidation,[46] and an increase in high-density lipoprotein.[43] On the other hand, heavy alcohol consumption is believed to be associated with adverse effects on the heart, including detrimental effects on blood pressure, fibrinolytic factors, and ventricular arrhythmias.[40] Another major adverse effect of alcoholic beverages is their carcinogenicity. The International Agency for Research on Cancer has classified cancers of the oral cavity, pharynx, larynx, esophagus, liver, colorectum, and female breast as being causally related to alcohol consumption.[47] For all of these sites except the liver and colorectum, increasing consumption is associated with higher risk.[47] The dose-response relationship is uncertain for cancers of the colon and rectum, whereas for the liver, it cannot be quantified because cirrhosis and other liver disorders that often precede liver cancer lead to decreases in consumption. Molecular and biochemical process for the causation of cancer by chronic exposure to alcohol include polymorphism in genes that encode enzymes responsible for ethanol metabolism, increased estrogen concentration, and changes in folate metabolism and DNA repair.[47]

In conclusion, our meta-analysis for men supports a curvilinear association between alcohol and all-cause mortality risk with a weak inverse association at lower intakes and a higher mortality risk for heavy consumption when consumption is measured over time (as opposed to at 1 time only). It also reaffirms how few cohort studies have captured alcohol intake for periods other than baseline and highlights the nonstandardized methods used to capture and analyze exposure even in the few studies that have measured intake longitudinally. It does not address whether a net gain in survival is seen in a particular age group and whether low-dose drinking in young adulthood renders any benefit in terms of reduced mortality risk in old age. According to Rehm and Sempos,[48] the curvilinear relationship between alcohol and mortality risk is, at least partly, dependent on the age groups observed (e.g., <60 vs. ≥60 years). Of the studies included in the present meta-analysis, only Goldberg et al.[24] reported relative risks for different age groups. Future studies measuring alcohol intake over time can add more to the existing literature by adopting standardized methods to capture intake and classify risk and by examining the effects of drinking for different periods of life in both men and women.